This paper presents a solution aimed at improving and expanding ensemble methods, where, depending on the properties of data coming from the robotic system, a model with the best quality indicators is selected, as well as the formation of multi-level structures that analyze incoming information flows and assign the most suitable model for processing sequences within the framework of solving the current problem. The proposed solution is based on the possibility of constructing hierarchies, when the upper-level model is used to assign the most effective lower-level model to a separate segment of the sample. Data coming from sensors of robotic systems often represent difficult to separate sequences. They contain many classes, heterogeneity of regions, different ranges of values. The results of the experiment show that on individual segments the values of the quality indicator of individual algorithms are better than when processing the entire sample as a whole. By implementing a function that analyzes quality indicators, it is possible to assign to a segment a model that has the best value on it. Training a single classifier rather than a group of complex classification models makes it possible to reduce computational costs in the context of limited robotic system resources.